diff --git a/src/asahi/compiler/agx_compile.c b/src/asahi/compiler/agx_compile.c
index a7228f0ae64..0a0ac758fc3 100644
--- a/src/asahi/compiler/agx_compile.c
+++ b/src/asahi/compiler/agx_compile.c
@@ -1489,6 +1489,10 @@ agx_emit_intrinsic(agx_builder *b, nir_intrinsic_instr *instr)
       return agx_get_sr_coverage_to(b, dst,
                                     AGX_SR_ACTIVE_THREAD_INDEX_IN_SUBGROUP);
 
+   case nir_intrinsic_load_active_subgroup_count_agx:
+      return agx_get_sr_coverage_to(b, dst,
+                                    AGX_SR_TOTAL_ACTIVE_THREADS_IN_SUBGROUP);
+
    case nir_intrinsic_reduce: {
       assert((instr->def.bit_size == 1 || instr->def.bit_size == 16 ||
               instr->def.bit_size == 32) &&
diff --git a/src/asahi/compiler/agx_validate.c b/src/asahi/compiler/agx_validate.c
index d2b34f5f70c..31b17a223b4 100644
--- a/src/asahi/compiler/agx_validate.c
+++ b/src/asahi/compiler/agx_validate.c
@@ -423,6 +423,8 @@ agx_validate_sr(const agx_instr *I)
    switch (I->sr) {
    case AGX_SR_ACTIVE_THREAD_INDEX_IN_QUAD:
    case AGX_SR_ACTIVE_THREAD_INDEX_IN_SUBGROUP:
+   case AGX_SR_TOTAL_ACTIVE_THREADS_IN_QUAD:
+   case AGX_SR_TOTAL_ACTIVE_THREADS_IN_SUBGROUP:
    case AGX_SR_COVERAGE_MASK:
    case AGX_SR_IS_ACTIVE_THREAD:
       return coverage;
diff --git a/src/compiler/nir/nir_divergence_analysis.c b/src/compiler/nir/nir_divergence_analysis.c
index 71f359db449..b8d4395b177 100644
--- a/src/compiler/nir/nir_divergence_analysis.c
+++ b/src/compiler/nir/nir_divergence_analysis.c
@@ -270,6 +270,7 @@ visit_intrinsic(nir_intrinsic_instr *instr, struct divergence_state *state)
    case nir_intrinsic_load_printf_base_identifier:
    case nir_intrinsic_load_core_id_agx:
    case nir_intrinsic_load_samples_log2_agx:
+   case nir_intrinsic_load_active_subgroup_count_agx:
       is_divergent = false;
       break;
 
diff --git a/src/compiler/nir/nir_intrinsics.py b/src/compiler/nir/nir_intrinsics.py
index 02e6653e64a..91b8e29f845 100644
--- a/src/compiler/nir/nir_intrinsics.py
+++ b/src/compiler/nir/nir_intrinsics.py
@@ -1850,6 +1850,10 @@ intrinsic("load_fep_w_v3d", dest_comp=1, flags=[CAN_ELIMINATE, CAN_REORDER])
 # Equivalent to popcount(ballot(true) & ((1 << subgroup_invocation) - 1))
 intrinsic("load_active_subgroup_invocation_agx", dest_comp=1, flags=[CAN_ELIMINATE])
 
+# Total active invocations within the subgroup.
+# Equivalent to popcount(ballot(true))
+intrinsic("load_active_subgroup_count_agx", dest_comp=1, flags=[CAN_ELIMINATE])
+
 # Like ballot() but only within a quad.
 intrinsic("quad_ballot_agx", src_comp=[1], dest_comp=1, flags=[CAN_ELIMINATE])